- Detach ref to new stub asset on overwrite when siblings share the asset
- Add optimistic mtime_ns guard in enrich_asset to discard stale results
- Normalize and validate output paths stay under output root, deduplicate
- Skip metadata extraction for stub-only registration (align with fast scan)
- Add RLock comment explaining re-entrant drain requirement
- Log warning when pending enrich drain fails to start
- Add create_stub_asset and count_active_siblings query functions
Amp-Thread-ID: https://ampcode.com/threads/T-019cfe06-f0dc-776f-81ad-e9f3d71be597
Co-authored-by: Amp <amp@ampcode.com>
Change _lock from Lock to RLock and move the start_enrich call inside the
lock-held block so that enqueue_enrich cannot interleave between clearing
_pending_enrich and starting the enrichment scan. This prevents a concurrent
enqueue_enrich from stealing the IDLE slot and causing the drained payload
to be silently dropped.
Add tests covering:
- pending enrich runs after scan completes
- enqueue during drain does not lose work
- concurrent enqueue during drain is queued for the next cycle
Amp-Thread-ID: https://ampcode.com/threads/T-019cfe02-5710-7506-ae80-34bf16c0171a
Co-authored-by: Amp <amp@ampcode.com>
Add ingest_existing_file() to services/ingest.py as a public wrapper for
registering on-disk files (stat, BLAKE3 hash, MIME detection, path-based
tag derivation).
After each prompt execution in the main loop, iterate
history_result['outputs'] and register files with type 'output' as
assets. Runs while the asset seeder is paused, gated behind
asset_seeder.is_disabled(). Populates job_id on the asset_references
table for provenance tracking.
Ingest uses a two-phase approach: insert a stub record (hash=NULL) first
for instant visibility, then defer hashing to the background seeder
enrich phase to avoid blocking the prompt worker thread.
When multiple enrich scans are enqueued while the seeder is busy, roots
are now unioned and compute_hashes uses sticky-true (OR) logic so no
queued work is silently dropped.
Extract _reset_to_idle helper in the asset seeder to deduplicate the
state reset pattern shared by _run_scan and mark_missing_outside_prefixes.
Separate history parsing from output file registration: move generic
file registration logic into register_output_files() in
app/assets/services/ingest.py, keeping only the ComfyUI history format
parsing (_collect_output_absolute_paths) in main.py.
Amp-Thread-ID: https://ampcode.com/threads/T-019cf842-5199-71f0-941d-b420b5cf4d57
Co-authored-by: Amp <amp@ampcode.com>
